Research Centre for International Economics Working Paper: … working paper... · 2014. 6. 9. · Research Centre for International Economics . Working Paper: 2014008 . Title of

Research Centre for International Economics

Working Paper: 2014008

Title of Paper The Penn Effect within a Country – Evidence from Japan

Authors’ List

Yin-Wong Cheung, City University of Hong Kong, and Eiji Fujii, Kwansei Gakuin University

Abstract

To control for product quality and exchange rate effects, we use the Japanese regional data to study the Penn effect – the positive relationship between price and income levels. Comparable with the evidence from international data, the Penn effect is significant in the Japanese prefectural data and driven mainly by the prices of nontradables. We draw upon studies of productivity and economic density to explain the positive price-income relationship, and find that the empirical economic density variables explain the variability of the Japanese prefectural (relative) prices quite well. © 2014 by Yin-Wong Cheung. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given to the source.

The Penn Effect within a Country –

Evidence from Japan

Yin-Wong Cheung

Eiji Fujii

Abstract

To control for product quality and exchange rate effects, we use the Japanese regional data to study the Penn effect – the positive relationship between price and income levels. Comparable with the evidence from international data, the Penn effect is significant in the Japanese prefectural data and driven mainly by the prices of nontradables. We draw upon studies of productivity and economic density to explain the positive price-income relationship, and find that the empirical economic density variables explain the variability of the Japanese prefectural (relative) prices quite well.

JEL Classifications: F31, F34, F36

Keywords: Agglomeration, Economic Density, Price and Income Relationship, Productivity

Differential, Tradables and Non-Tradables.

Acknowledgments: We are grateful to Francis Teal, the managing editor, and two anonymous referees. Their comments and suggestions greatly improved the manuscript. We also thank Michael Funke, Jan Fidrmuc, Stephan Hansen, and participants of the CESifo Macro, Money and International Finance Conference for their comments and suggestions on earlier drafts of the paper. Fujii gratefully acknowledges the financial support of the Japan Society for the Promotion of Sciences (Grant-in-Aid for Scientific Research 20530263, 25285087). Corresponding addresses: Yin-Wong Cheung: Department of Economics & Finance, City University of Hong Kong,

Hong Kong. E-mail: [email protected] Eiji Fujii: School of Economics, Kwansei Gakuin University, Hyogo, Japan. E-mail:

[email protected].

1

1. Introduction

The Penn effect refers to the robust empirical positive association between national

price levels and real per capita incomes that is documented by a series of Penn studies

including Kravis et al. (1978), Kravis and Lipsey (1983, 1987), and Summers and Heston

(1991). The accumulated evidence attests that compared with poor countries, rich countries

tend to have higher price levels. The positive association between income and price levels is

considered a fundamental fact of economics (Samuelson, 1994) and conventional wisdom in

international economics (Bergin, 2009).

Most studies analyzing the price-income relationship have relied on data derived from

the surveys conducted by the International Comparison Program (ICP). Since its

establishment in 1968, the ICP has conducted periodic surveys on national prices. The survey

results are used to construct internationally comparable price indices and national output data.

The ICP has expanded the country sample and product coverage, and improved the survey

methodology and data processing procedure over the last few decades to enhance the

reliability of the survey.1 Despite the ICP’s continuing efforts, it remains a daunting task to

aggregate and compare prices of vastly dissimilar products from countries of different

economic characteristics, and over time (Deaton and Heston, 2010).

Several factors contribute to the difficulty of constructing internationally comparable

data. For instance, national price level comparison becomes quite tricky, if not infeasible,

when countries differ substantially in their output structures and consumption patterns. These

differences are not uncommon between countries at different stages of development and with

different cultural backgrounds. For a given product, a meaningful comparison of its prices in

different countries has to control for its quality attributes; actual or perceived. It is quite

difficult to quantify quality differentials for nontradables such as locally provided services.

To alleviate the concerns regarding data incompatibility, the current study investigates

the price and income relationship using Japanese regional data. The Japanese data offer a few

desirable features. For instance, the price data are collected from national surveys that are

designed to cover products of the same quality and quantity attributes across locations.

Further, availability of disaggregated price series enables us to examine behavior of sectoral

1 The changes in the survey setup led to considerable data revisions that have profound implications for estimates of growth rates, growth determinants, poverty measures, and inequality indicators; see, for example, Johnson et al. (2009), Milanovic (2009), Chen and Ravallion (2010a, 2010b), Ciccone and Jarocinski (2010), and Ponomareva and Katayama (2010). On the Penn effect, Cheung et al. (2009) and Fujii (2013a) illustrate that, although the positive price-income relation survives data revisions, the magnitude of the estimated income effect has been noticeably changed.

2

prices. The income data are compiled under identical accounting and tax systems. While

consumption bundles may still differ by regions, the degree of consumption homogeneity

within Japan is arguably higher than that across countries. In addition, the intra-Japanese

comparison is free from the exchange rate volatility effect that sometimes inflicts

cross-country comparison.

There are no legal restrictions on either goods or factor flows between regions within

Japan. The condition has two implications. First, unlike the international cases, the

intra-national price-income relationship is not affected by differences in policies on trade and

factor flows. Thus, it is relatively easy to interpret results based on intra-national data.

Second, while free trade enhances goods price convergence, free factor movement can

generate equalizing pressure on prices and qualities of local services across regions.2 Of

course, the net effect of goods and factor flows depends on the de facto mobility rather than

de jure restrictions because implicit barriers and frictions exist even within a country.3 Given

these differences, it is of interest to compare intra-national estimates to international ones.

In addition to data features, the current study takes an alternative approach and draws

upon theories on productivity and economic density (Ciccone and Hall, 1996) to explain the

intra-Japan Penn effect. Specifically, the economic density that significantly affects labor

productivity is used to explain the price-income relationship observed in the Japanese data.

Because good quality regional productivity data in Japan are scarce, the empirical economic

density measure offers a good alternative to assess the relevance of the well-known

productivity differential effect à la Balassa (1964) and Samuelson (1964).

To anticipate the results, we find that, across the Japanese regions, the price and

income levels are significantly positively associated with each other. That is, the Penn effect,

commonly documented with international data, is also a staple feature of the intra-Japanese

data. Our attempts to account for the intra-Japan price-income relationship offer some

evidence that the Balassa-Samuelson (B-S) effect retains its relevance in the intra-national

context. First, as implied by the usual B-S argument, the positive price-income association is

driven essentially by prices of nontradables rather than those of tradables. Further, we find

that price differentials across regions are significantly determined and explained by

2 In principle, households can migrate to areas where quality services are cheaper ceteris paribus. Private service-providers can also choose their locations. The adjustment across regions may take place in terms of either price, quality or both. 3 For instance, factors such as uneven distribution of industrial locations, climatic differences, and family ties within a region may work as implicit barriers for labor flow. Also, non-negligible transport cost generates market friction for flow of products.

3

differences in regional economic density characteristics. Under the assumption that economic

density is a proxy of productivity, our finding is consistent with the presence of the B-S effect

in the Japanese data. Nonetheless, the effect of economic density could have a broader

interpretation than the B-S effect because, in addition to the supply-side effect, economic

density can affect relative prices of nontradables via the demand-side channel.

The remainder of this paper is organized as follows. Section 2 provides a brief

background discussion, describes the data, and defines the empirical variables. Section 3

presents the empirical price-income relationship between the Japanese prefectures. The

results are then compared to those obtained from international data. Section 4 investigates the

implications of tradability and productivity differentials using data on different product

categories. In section 5, we evaluate the role of economic density in explaining the

price-income relationship within Japan. Some concluding remarks are offered in Section 6.

2. Preliminaries

2.1 Aggregate Price Level

At the risk of over-simplification, suppose region j ’s aggregate price level in period t,

in logarithm, is given by

, , , , , (1 )j t N j t T j tp p p , (1)

where , ,N j tp is the price of nontradables, , ,T j tp is the price of tradables, and defines the

weights of the price components. Similarly, the logged aggregate price level of region j* is

*, , *, , *, * (1 *)j t N j t T j tp p p . (2)

To facilitate comparison, the exchange rate of the two regions’ currencies is used to convert

the two prices ,j tp and *,j tp into the same unit. For our data, the regions are within Japan.

Thus, the exchange rate is fixed at unity, and ,j tp and *,j tp can be directly compared.

Two additional assumptions commonly imposed are a) the two prices use the same

weight; that is, = * , and b) the prices of tradables are the same across different regions

so that , ,T j tp = , *,T j tp . Under these assumptions, the conventional sectoral productivity

differential argument suggests that a less productive and, hence, lower income region will

have a lower price of nontradables and a lower aggregate price level.

The simple setting outlined above highlights a few controversial issues that inflict

cross-country price comparison. In addition to exchange rate volatility, the ability to compare

prices is impeded by the fact that aggregate price levels are not necessarily compiled using an

4

identical methodology. Further, prices of tradables are not necessarily the same across

countries (Engel and Rogers, 1996). To further complicate the situation, national aggregate

price levels comprise prices of individual products that have heterogeneous, rather than

homogeneous, qualities across countries. The quality difference does not only create a wedge

between prices of nontradables but also between prices of tradables (Imbs et al., 2010).

2.2 Price Data

We use data from forty-seven prefectures in Japan. The Japanese prefectures are

geographically defined administrative units largely corresponding to, say, the States in the US.

Specifically, we use the Regional Difference Index of Consumer Prices (RDICP) provided by

the Statistical Bureau of the Ministry of Internal Affairs and Communications. The sources of

the price and other variables used in the following empirical exercise are listed in the

Appendix. Instead of absolute price levels, these RDICP series report prefectural consumer

price levels relative to the price level of the Tokyo central area, which comprises twenty-three

districts of Tokyo prefecture. With central Tokyo as the common reference, the RDICP data

allow us to gauge the price differentials of these forty-seven Japanese prefectures.

The RDICP series are derived from the price information collected by the Statistical

Bureau’s retail price survey. The survey records retail prices of products and services that are

quite precisely defined. Examples of product and service descriptions include “hen eggs

(color: white, size L, sold in pack of 10),” “men’s undershirt (short sleeves, knitted, white,

100% cotton, [size] around the chest 88-96cm/MA (M), white, ordinary quality, excluding

specially processed goods),” and “permanent wave charges (including shampoo, cut, blow or

set) for short hair.” In many categories, product brands are specified to ensure that prices are

recorded for identical products. For instance, ice cream prices collected by the survey are

prices of “Häagen-Dazs vanilla (by Häagen-Dazs Japan), 120ml”. The specificity of product

definition enhances price comparability and minimizes the role of product heterogeneity in

explaining price differentials across prefectures.4

It is noted that the consumption tax is completely harmonized across all regions in

Japan. In addition, the consumption patterns across these Japanese prefectures are arguably

more homogeneous than those across countries. Thus, the prefectural price differentials are

less subject to the effects of differential taxes and dissimilar consumption patterns. In sum,

4 However, the perception of heterogeneity may be induced by factors not controlled for in the survey, including the characteristics of the store in which the products are sold.

5

the use of these Japanese price data alleviates some of the measurement and data

incompatibility issues raised in the previous subsection.

2.3 Basic Empirical Variables

The price variable used in this study is the deviation from the prefectural average.

Specifically, a prefecture’s aggregate price level relative to the average of prefecture price

levels, in logs, is measured by

47/lnln 471 ,,, j tjtjtj RPRPq , (3)

where tjRP , is prefecture j’s RDICP in period t. A similar normalization procedure is also

applied to the per capita income data derived from information on real gross prefectural

income and prefectural population provided by the Statistical Bureau. The prefectural real per

capita income relative to the average of prefecture data, in logs, is thus given by 47

, , , , ,1ln( / ) ln( / ) / 47j t j t j t j t j tjy Y H Y H , (4)

where ,j tY and ,j tH denote prefecture j’s real gross prefecture income that includes net

factor payments from other prefectures, and population in period t, respectively.5

Due to data availability, annual data from 1996 to 2008 are considered. The time

averages of tjq , and ,j ty are listed in Table A-1 of the Appendix. The relative aggregate

price level ranges from about five percentage points below (-0.049, Okinawa) to more than

eight percent above (0.085, Tokyo) the average. Inter-prefectural per capita income

differentials are far more substantial. As shown in the far right column of Table A-1, the

time-average of per capita income relative to the average ranges from -0.29 (Okinawa) to

0.50 (Tokyo). The income variation helps identify the income effect on prices.

3. The Penn Effect within Japan

For each year, we estimate the Penn effect within Japan using the canonical

cross-sectional bivariate specification

, , ,j t j t j tq y . (5)

5 The income and population data are available for Tokyo prefecture but not for Tokyo central area. In addition to the twenty-three districts in the center, Tokyo prefecture includes twenty-six cities, five towns, and eight villages. The RDICP uses Tokyo central area as the benchmark. The normalization procedure adopted by (3) and (4) ensures the price and income data are comparable.

6

The time profile of the slope coefficient estimate ̂ and its p-value obtained from

year-by-year cross-sectional regression are depicted in Figure 1.6 The estimated effect of

income on price is significantly positive throughout the sample period. Even though the

year-by-year estimates display some variation, the parameter stability tests indicate that these

estimates are not statistically different from each other. The Penn effect is a robust empirical

feature of the Japanese data.

How does the Penn effect within Japan compare to the one documented using

international data? Figure 2 plots the year-by-year income effect estimates, ̂ s, obtained

from the corresponding data downloaded from the Penn World Table (version 7.0) and the

World Development Indicator (January 2012), together with those from the Japanese data.

The regressions based on international data use the US as the reference country, and include

all countries with available observations. Although both PWT and WDI are derived from the

same ICP 2005 survey information, they adopt different approaches in their data compilation

methods. Thus, the estimates, ̂ s, from these international data are not the same. Further,

while the Japanese regional data are CPI-based price data, the PWT and WDI data are

GDP-based data. These differences should be considered in comparing these ̂ -estimates.

In line with the extant literature, the Penn effect is identified in the two international

datasets. As the plots indicate, the income effects exhibited by these international data are

more substantial than the one by the Japanese data; that is, compared with cross-country

behavior, the change in the income within Japan tends to induce a smaller change in the price

level. Further, the evolution of the Japanese ̂ estimates is discernibly different from those

of the other two ̂ -estimate series. Aside from these differences, however, the three datasets

unanimously exhibit significant positive income effects on price levels, which is a defining

signature of the empirical Penn effect. In sum, despite the differences in data compilation and

construction methods, the Penn effect appears a prevalent phenomenon in both the

cross-country and intra-Japan data.

The results thus far indicate that price and income levels of Japanese prefectures are

positively related to each other – a relationship that is also revealed by cross-country data.

The Penn effect in Japan is qualitatively similar though not quantitatively identical to the

cross-country Penn effect.

6 Since both the dependent and independent variables are deviations from the respective sample averages, the intercept is zero by construction. Thus, the constant estimates are insignificantly different from zero and not reported for brevity.

7

4. Prices of Nontradables and Tradables

A common explanation of the positive income effect on prices draws on the different

price behaviors of nontradables and tradables and the difference in sectoral productivities

(Balassa, 1964; Samuelson, 1964). In this section, we examine the implications of the

tradable-nontrable dichotomy for the intra-Japan positive price-income association.

Following (1) and (2) in Section 2.1, the relative price level of two regions, in the

presence of a perfectly fixed exchange rate and under the assumption of = * , is given by

*,j tp – ,j tp = ))(1()( ,,*,,,,*,, tjTtjTtjNtjN pppp . (6)

Further, if the prices of tradables are the same under the usual arbitrage argument and in the

absence of border effects (Engel and Rogers, 1996), then the relative price level is merely

proportional to the relative price of nontradables. In this case, the price-income relationship

essentially reflects the link between prices of nontradables and income levels. Of course,

prices of tradables are not necessarily identical across regions. Nevertheless, if prices of

tradables, compared with prices of nontradables, are more likely to converge, then the income

effect should be more pronounced on prices of nontradables than tradables.

The implication of the degree of tradability is examined using the specification:

, , , , ,k j t k k j t k j tq y , (7)

where , ,k j tq is the relative price index of product category k in region j at time t derived

based on the procedure defined by (3) using data on the corresponding sub-price index of the

RDICP. An overarching issue is how to determine which product category is tradable and

which is nontradable. The dichotomy between nontradables and tradables is a convenient

device in theoretical analyses. In reality, however, most if not all consumer products contain

both non-tradable and tradable components. Products are neither strictly tradable nor

nontradable, but have different degrees of tradability. Thus, as an empirical classification

scheme, the dichotomy of nontradables and tradables is quite restrictive. With the caveat in

mind, we use data on price indexes of different product categories to assess the role of

tradability.

The list of prefecture product-category price indexes is given in the Appendix, Table

A-2. The data on these prefectural sub-indexes of the RDICP are published every five years

and available only for 1997, 2002, and 2007 during the sample period under consideration.

The year-by-year income effect coefficient estimates, k̂ s, from product-category-specific

8

price index data are presented in Table 1 and graphed in Figure 3. The estimates from the

prefectural consumer price level data are included for comparison purposes. The last column

of Table 1 additionally reports the estimates by pooling the data from the three sample years.

A few observations are in order.

First, in view of the disaggregated prices listed in Table A-2, the category “services”

is commonly conceived to be less tradable than the category “goods.” Indeed, in all years, the

income effect coefficient estimate of the “services” category is larger than that of the “goods”

category. Further, the former is statistically significant while the latter is not. The results are

in line with the notion that the Penn effect is driven by nontradables.

Second, the “goods” group consists of products that are not equally tradable. Even

though data on “goods” have an insignificant k̂ estimate, the sub-categories “agricultural

& aquatic products” and “fresh agricultural & aquatic products” display a significant

price-income relationship for 1997 and 2002.

The results could be attributed to their perishable characteristic. These two perishable

sub-categories are likely to have a lower degree of tradability than, say, industrial products,

and exhibit the Penn effect like a nontradable product. According to k̂ estimates, the group

of “fresh agricultural & aquatic products” yields a stronger income effect than the group of

“agricultural & aquatic products.” The improvements in transportation and storage

technologies enhance the tradability of perishable products, and thus, weaken the Penn effect

in the 2007 sample. It is also noted that, the income effect exhibited by the “CPI excluding

fresh foods” category is weaker than the one by the “CPI” data.

For the 1997 and 2002 regressions, the income displays no significant effect on the

prices of the subcategory “industrial products,” which are in general nonperishable and

perceived to be highly tradable. Nevertheless, the income effect became significant in 2007.

While we do not have a definitive explanation for the switch in significance over time, the

results are suggestive of the possibility that the degree of tradability can vary not only across

product categories but also over time.

The income effect estimate for “publications” is insignificant. Products in this

sub-category including books, magazines and newspapers, tend to have nation-wide listed

prices. Such practices limit the variability of regional prices, and make the product prices

unresponsive to income changes.

The significantly negative coefficient estimates obtained for the “electricity, gas &

water charges” sub-category deserves a comment. While this sub-category is included under

9

the heading of “goods,” its components are mostly utilities, and their prices are subject to

local administrations and regulations. Water, for example, is usually supplied by municipal

governments, whereas electricity and gas are by monopolistic firms in geographically defined

markets. Thus, these prices are less likely subject to the usual arbitrage forces.

The culprit of negative coefficient estimates could be the economies of scale effect

underlying the provision of public services and utilities. For instance, if there is a large fixed

cost of providing electricity and gas services, the average utility charge can decline with the

population and the number of households using the services. In our prefectural dataset, the

population size is positively correlated with the income level. Thus, compared with less

affluent prefectures, more affluent prefectures can have lower utility charges.7

Third, by the same token, products within the “services” category have different

degrees of tradability. The “public services” and “general services” sub-categories have

starkly different income effects. The “public services” include public housing, medical and

welfare, communication and transportation, and educational services. These services are

generally not tradable between prefectures, and their prices tend to be regulated. Our results

show that the prices of “public services” are income insensitive. On the other hand, the price

of privately provided “general services” exhibits a highly significant and large income effect.

That is, the price of “general services” tends to be higher where real income is higher.

The two subgroups of “general services;” namely, the “private house rent” and “eating

out” differ substantially in their income coefficient estimates. While both k̂ s are statistically

significant, the estimated income effect on the subgroup of “private house rent” is much

stronger than on the subgroup of “eating out” prices. One speculation is that prevalence of

chain-stores in the eating out industry weakens the income effect on its price.

The results pertaining to the pooled data are essentially the same as those discussed

above. The main exception is that the income effect is significant for the “goods” category,

albeit the magnitude is relatively small. The significance can be attributed to the increase in

estimation efficiency due to an increase in sample size, and the fact that not all items under

the category are tradable.

Overall, the prices of products with different degrees of tradability respond differently

to income. The Japanese prefectural data yield results that confirm the common wisdom;

income tends to have a larger impact on prices of nontradables than on prices of tradables. An

7 Indeed, on the average, the prefectural price of “electricity, gas & water charges” is lower in prefectures with a larger population size. These results are available upon request.

10

implication is that the observed positive association between price and income levels is

largely attributable to nontradables. Of course, the interpretation is subject to the usual caveat

that we do not have a precise measure of the degree of product tradability.

5. Accounting for the Intra-Japan Penn effect

5.1 Productivity and Economic Density

The difference in the relative sectoral productivity is the basis of the B-S hypothesis,

which is a long-standing explanation of the international price-income relationship. 8

Empirical studies on the productivity differential effect have evolved over time. One key

issue is the choice of productivity measure, which has varied from per capita gross national

product to some specifically constructed measures of sectoral productivity.9 Further, some

studies consider (average) labor productivity while others use total factor productivity.10 The

comparison of levels of national productivity is further complicated by differences in

methods of reporting economic data and in data quality.

Productivity is not directly observable in general. While empirical measures of

productivity are routinely used, there are concerns about how well the empirical measures can

capture the notion of productivity used in theoretical models. Indeed, because of the paucity

of the Japanese prefectural productivity data, we explore alternative proxies for

productivity. 11 Specifically, we consider the proxies that are motivated by studies on

economic density (Carlino and Voith, 1992; Ciccone and Hall, 1996), and the related

agglomeration argument (Henderson, 1974; Krugman, 1991; Glaeser, 2008).

Economic density refers to the intensity of labor, human capital, and physical

investment relative to the physical space. Ciccone and Hall (1996) note a few channels

through which economic density can affect the level of productivity in a locality:

rising-by-distance transport costs from one production stage to the next; externalities

associated with physical proximity of production; and a high degree of beneficial

8 Other explanations advocated in the literature include the factor-intensity and factor- endowment approach (Kravis and Lipsey, 1983; Bhagwati, 1984), and the non-homothetic demand structure approach (Bergstrand 1991). 9 For different choices of productivity measures, see, for example, Balassa (1964), Officer (1976), Hsieh (1982), Asea and Mendoza (1994), De Gregorio et al. (1994), Canzoneri et al. (1999), Chinn (2000), and Kakkar (2003). 10 For example, Marston (1987) and Canzoneri, Cumby, and Diba (1999) use average labor productivity, while Asea and Mendoza (1994) and De Gregorio et al. (1994) use total factor productivity. 11 Prefectural data constraints are quite severe. We explored the use of proxy measures for labor productivity that are constructed using incomplete prefectural data on sectoral value added and employment. The results obtained from these noisy measures – not reported for brevity but available from the authors – are mostly insignificant.

11

specialization possible in areas of dense activity. One prediction of their analyses is that a

locality with a higher average employment density and a higher inequality of employment

density will have a higher level of productivity. There is a caveat however. Increasing

economic density generates not only agglomeration effects that raise productivity, but also

congestion effects. If congestion effects outweigh agglomeration effects, then a high density

area will have a low level of productivity. It is an empirical matter to determine which of the

two effects prevails. Using the US state and county data, the authors find that a rise in

employment density leads to a significant increase of average labor productivity.12

5.2 Empirical Exploration

Because prefectural employment density data are not available, we capture the

economic density effect using a) the prefectural population density, and b) the population

density of the most agglomerated areas within a prefecture. The second variable is based on

the population data of densely inhabited districts (DIDs), which are districts that have more

than four thousands inhabitants per square kilometer. The two variables correspond to the

average employment density and the employment density inequality in Ciccone and Hall

(1996). To control for differences between employment and population data, we include data

on unemployment rates in our analysis.13

The economic density effects are examined using the regression

, 1 , 2 , 3 , ,j t j t j t j t j tq density DID UE , (8)

where tjdensity , and ,j tDID are, respectively, the number of inhabitants per square

kilometer in prefecture j and in DIDs in the same prefecture at time t. Similarly to the price

and income variables, the density variables are in logarithmic terms and expressed as

deviations from their respective averages. ,j tUE is the unemployment rate deviations from

the average at time t.

The data on population density and DIDs are available only every five years, and for

1995, 2000, and 2005 during our sample period. Thus, 1995, 2000, and 2005 density data are

paired up with the corresponding 1997, 2002, and 2007 disaggregated price data. We estimate

(8) using a) year-by-year cross-sectional data, and b) the pooled data while allowing for

12 In a similar vein, Carlino and Voith (1992) find that total factor productivity across the U.S. states increases with the level of urbanization, and Glaeser and Maré (2001) report evidence on labor productivity and wages. 13 An implicit assumption is that unemployment rates, while varying across prefectures, are constant across sectors within each prefecture. Ideally, one should use employment data that are sector-and-prefecture-specific to calculate employment density. However, these data are not available.

12

year-specific intercepts. To conserve space, we focus on pooled data estimations, and present

the year-by-year regression results in the appendix.

According to the economic density reasoning, we expect the two density variables to

have positive effects. Assuming that the unemployment rate is constant, a prefecture with a

high population density (represented by ,j tdensity ) has a high average employment density

that implies a high level of productivity, a high level of income, and hence, a high general

price level. When the average density is held constant, ,j tDID reflects the extent of density

inequality within a prefecture since it captures essentially the district-specific highest density.

A theoretical prediction is that, for a given level of average density, density inequality within

a prefecture intensifies the overall effect and boosts the level of productivity. Thus, ,j tDID

is also expected to exert a positive effect on prices. We expect ,j tUE to have a negative

effect; a high unemployment rate means a low effective employment density, ceteris paribus.

The pooled estimates are summarized in Table 2. As column 1 presents, the average

density and DID density variables are jointly significant and positive. The results are in

accordance with Ciccone and Hall (1996); that is, the two density variables convey different

types of information about price levels. Under the presumption that the economic density

variables are proxies for productivity levels, the finding lends support to the link between

productivity and price levels. That is, a prefecture with a high level of economic activity and

economic activity inequality tends to have a high general price level. The unemployment rate

variable has a significantly negative coefficient estimate as expected. The combined

explanatory power of these three variables is quite high at the 60% mark.

Do the density effects fully account for the intra-Japan Penn effect? Column 2

evaluates the marginal effect of the income variable. When the income variable is added to

(8), the DID variable retains its significance while the average density variable becomes

statistically insignificant. The results indicate that the price information content of the income

variable dominates that of the average density, and is not identical to the DID density variable.

In terms of marginal explanatory power, inclusion of the income variable leads to

improvement of the adjusted R-squared estimates albeit relatively small.

Columns 3 to 5 compare the individual effects of the density and income variables.14

Individually, they all display highly significant price effects. The DID variable coupled with

14 To capture the employment density effect, the average density and DID density variables are accompanied by the unemployment rate variable.

13

the unemployment rate offers the highest explanatory power among these three specifications,

while the income variable offers the lowest.

Table 3 presents the correlation coefficients between the explanatory variables. The

average density variable exhibits sizable correlations of .77 and .58, respectively, with the

DID density and income variables. The correlation between the DID density and income

variables is lower at .33. The insignificance of the average density effect under column 2 of

Table 2 is possibly due to its relatively high degrees of correlation with the other two

variables. While there is overlapping in price informational contents, the combined

specification considered under column 2 shows that the empirical Penn effect cannot be

entirely explained by economic density factors.

The year-by-year results, summarized in Table A-3 in the appendix, convey a very

similar message. In general, the average density and DID density variables are jointly

significant. The significance of the average density variable is weakened in the presence of

the income variable, while that of the DID density variable is not. The relative individual

effects and explanatory powers are comparable to those presented in Table 2.

Similar to studies of the B-S effect, we assess the roles of economic density in

explaining the relative price of nontradables to tradables using the regression specification:

, , , , 1 , 2 , 3 , ,( )N j t T j t j t j t j t j tq q density DID UE , (9)

where subscripts N and T denote nontradables and tradables sectors, respectively.

We note that the two density variables in (9) are not sector-specific and, hence, are not

direct measures of the sectoral productivity differentials. However, if cross-prefecture

productivity differences in nontradables sectors are negligible as assumed under the B-S

hypothesis, then the economic density variables reflect differences in the productivity levels

of tradables sectors. Under this assumption, the productivity differential effect argument

implies that 01 and 02 . The assumption and its limitations need to be taken into

account when interpreting the empirical results.

The measure of the relative price of nontradables to tradables is based on the relevant

sectoral price indices. Based on the results in section 4, we use the price index of “general

services” as our proxy for the price of nontradables , ,N j tq , and the price indices of “industrial

products” and “goods” as two alternative proxies for the price of tradables.

In Table 4, panels A and B summarize the pooled estimates for the relative prices

based on “industrial products” and “goods”, respectively, as tradables. As displayed in

column 1, the density variables jointly attain the expected highly significant positive

14

coefficient estimates. Further, the adjusted R-squared estimate suggests that the density

variables along with the unemployment rate variable explain about 60% of the variations in

the relative price of nontradables to tradables between the Japanese prefectures.

Some previous international studies report significant demand side effects in

modelling the relative nontradable prices. For instance, De Gregorio et al. (1994) use per

capita income as a demand shifter, rather than as a productivity proxy. The results of

including the income variable in (9) are presented in column 2. The influences of the

presence of the income variable are similar to those revealed by the general price level

estimation (column 2 of Table 2); that is, the average density variable becomes insignificant

while the DID density variable is still significant. The marginal contribution of the income

variable to the overall explanatory power is relatively small in magnitude, albeit the estimate

of its coefficient is statistically significant.15

Columns 3 to 5 indicate that the economic density and income variables are

individually significant with the expected positive effect on the relative prices. The results do

not depend on the choice of the proxy for tradable price index. The adjusted R-squared

estimates suggest that the economic density variables, as compared to the income, have a

higher degree of explanatory power. Overall, the results of the relative price regressions

(Table 4) are qualitatively comparable to those of the general price regressions (Table 2).16

5.3 Discussion – Economic Density and Prices

In the previous sub-section, we found that the two proxies for economic density have

significant positive effects on the Japanese prefecture relative price of nontradables to

tradables. Under the presumption that economic density is related to productivity (Ciccone

and Hall, 1996), the finding implies that price differentials are related to productivity

differentials. If productivity gains tend to concentrate in the tradables sector, then the

inference could be extended to the context of sectoral productivity differentials.

While the economic density variables capture productivity effects, the mechanism

through which economic density affects the relative price of nontradables to tradables is not

15 We also considered the real government expenditure share of GPI as an additional control for demand side effects (Froot and Rogoff, 1991; De Gregorio et al., 1994). However, its effect is not found significant in any case. Further, the government expenditure variable exhibits a strong negative correlation with per capita income (-.81 by the pooled sample), tending to mask the effect of the income variable. To conserve space, these additional results are not reported but are available upon request. 16 With some minor yearly variations, the year-by-year estimation results summarized in Tables A-4-1 to A-4-3 in the appendix are qualitatively similar to the results from pooled data reported in the text.

15

identical to the one underlying the standard B-S explanation. One possible transmission

channel is as follows. The agglomeration of economic activities promotes economic

opportunities and induces productivity gains in the locality. The productivity gains, in turn,

impose pressure on wages in the sector of tradables (Glaeser and Maré, 2001) and, assuming

inter-sectoral labor mobility, in the sector of nontradables. Higher wages attract workers to

the region, resulting in an increase in the population density. The growing population density

in turn propels the demand for locality-specific nontradables including housing and other

locally-provided services. Prices of nontradables experience an upward pressure because

there is an increase in demand and in input costs including rents.

Under this setup, even though one observes co-movement between income and the

general price level, the prices of nontradables are affected by both supply and demand

factors.17 Under the B-S framework, nontradables are produced using labor and capital that

have an elastic supply. The economic density approach, however, recognizes the possible role

of inelastic supply of land in affecting the prices of nontradables.18

The increase in production costs that include land prices and rents will provide

incentives to improve labor productivity in the nontradables sector. The economic

propagation mechanism underlying the economic density exposition, thus, suggests that the

variation in population density can be a proxy for factors that drive nontradables prices, and a

locality’s population density should be related to its nontradables productivity and income, in

addition to its general price level. Both demand and supply factors can be in action. Thus, the

empirical economic density effect could be driven by both demand and supply forces, and not

necessary the same as the supply-side-driven B-S effect. Our empirical findings, nevertheless,

show that the Penn effect represented by the income effect is not entirely explained by

economic density.

6. Concluding Remarks

The Japanese prefectural data are employed to investigate the price-income

relationship, which is known as the Penn effect. Compared with most cross-country analyses,

one advantage of using the Japanese data is that the empirical finding is less likely to be

affected by product quality differentials as the Japanese prices are measured rather precisely

and consistently. In addition, the use of the intra-national data effectively eliminates nominal

17 It is implicitly assumed that different regions have a similar level of non-labor income. 18 In the B-S model, inter-regional perfect mobility and, hence, elastic supply of capital play an essential role in deriving the result that the regional prices of nontradables are determined solely by supply factors.

16

exchange rate volatility that could distort the observed link between price and income levels

across locations. Thus, our investigation offers a less ambiguous way to infer the prevalence

and robustness of the Penn effect, which is commonly documented in cross-country studies.

Our empirical results reveal that the widely documented cross-country positive

price-income association is also a staple feature of the Japanese data. The observed

intra-Japan Penn effect resembles the international one in that it is driven mainly by the

behavior of the prices of nontradables rather than tradables.

In accounting for the intra-Japan Penn effect, we draw on studies of economics of

agglomeration and, specifically, implications of economic density on productivity. The notion

of economic density offers an alternative way to infer the link between productivity, income,

and price levels that is different from the usual B-S interpretation.

We find strong evidence that the relative prices of nontradables to tradables in Japan

are driven by regional economic density characteristics. The empirical economic density

variables are found to possess large incremental explanatory power. For most specifications

considered, the economic density variables explain about 60% or more of the variation in the

Japanese prefectural price differential of nontradables and tradables. Although the income

effect on (relative) prices is weakened in the presence of economic density variables, the

intra-Japan Penn effect cannot be totally explained by the productivity-cum-economic-density

nexus. In other words, the income and density variables we adopted contain some

non-overlapping information about prices. Informational contents of per capita income

differentials can be broad and multifaceted. A higher income level can be simultaneously an

outcome of a higher productivity level and a source of a greater demand for goods and

services. Since our economic density variables also reflect both demand and supply factors, it

is possible that they cover different facets of demand and supply forces that drive price

variability. Our results warrant a future study on the roles of the B-S hypothesis and the

agglomeration and economic density approach in explaining the Penn effect.19

19 As illustrated in Appendix A3, the income effect on prices among rich prefectures is different from the one among poor prefectures. It is another area warrants additional analyses.

Appendix

Appendix A1. Data Sources

The Japanese regional data are obtained from the Regional Statistics Database of the

Statistics Bureau, Ministry of Internal Affairs and Communications. The international data

are obtained from the PWT 7.0 and the World Development Indicators database (January

2012).

Appendix A2. Additional Tables Table A-1. Relative consumer prices and per capita gross prefectural incomes 1996-2008

id Prefecture ,.jq ,.jy

1 Hokkaido 0.014 -0.065 2 Aomori 0.001* -0.182 3 Iwate -0.007 -0.097 4 Miyagi -0.003* -0.028 5 Akita -0.024 -0.115 6 Yamagata 0.006* -0.080 7 Fukushima -0.010 0.030 8 Ibaraki -0.009 0.041 9 Tochigi 0.004 0.082 10 Gunma -0.026 -0.002* 11 Saitama 0.025 0.023 12 Chiba 0.006 0.043 13 Tokyo 0.085 0.498 14 Kanagawa 0.070 0.152 15 Niigata 0.006 0.011 16 Toyama -0.003* 0.138 17 Ishikawa 0.006* 0.092 18 Fukui -0.004 0.045 19 Yamanashi 0.000* -0.035 20 Nagano -0.013 0.047 21 Gifu -0.013 -0.013 22 Shizuoka 0.030 0.123 23 Aichi 0.026 0.248 24 Mie -0.008 0.103 25 Shiga -0.008 0.162 26 Kyoto 0.035 0.019 27 Osaka 0.054 0.127 28 Hyogo 0.022 0.037 29 Nara -0.002* -0.092 30 Wakayama 0.003 -0.112 31 Tottori -0.021 -0.089 32 Shimane 0.007 -0.097 33 Okayama 0.010 0.027 34 Hiroshima -0.012 0.062 35 Yamaguchi -0.015 0.033 36 Tokushima -0.028 -0.031 37 Kagawa -0.018 -0.020 38 Ehime -0.039 -0.088 39 Kochi -0.015 -0.233 40 Fukuoka 0.002* -0.057 41 Saga -0.023 -0.111 42 Nagasaki 0.014 -0.209 43 Kumamoto -0.018 -0.177 44 Oita -0.022 -0.010* 45 Miyazaki -0.047 -0.186 46 Kagoshima -0.007 -0.194 47 Okinawa -0.049 -0.285

Notes: The entries are 1996-2008 averages of relative consumer prices and real per capita gross prefectural incomes. For each year, prefectural price and income levels are measured in logged deviations from their all prefecture averages. The entries with “*” are statistically not different from zero at the 5 % significance level.

Table A-2. Product categories of disaggregated regional price indexes

Goods Agricultural & aquatic products

Fresh agricultural & aquatic products Industrial products Electricity, gas & water charges Publications

Services Public services General services

Private house rent Eating out

Table A-3. Density effects on the price level by year

A. 1997 1 2 3 4 5 Average density .011**

(.004) .005 (.006)

.024** (.004)

DID density .076** (.016)

.075** (.014)

.113** (.016)

Unemployment rate

-.008* (.003)

-.005 (.004)

-.005 (.004)

-.008* (.004)

Income .055† (.029)

.138** (.029)

Adjusted R2

.685 .703 .548 .626 .425

B. 2002 1 2 3 4 5

Average density .008* (.004)

.005 (.006)

.017** (.004)


.053** (.014)

.080** (.015)

Unemployment rate

-.010* (.003)

-.007 (.005)

-.007 (.004)

-.010* (.004)

Income .027 (.038)

.100** (.029)

Adjusted R2

.514 .512 .417 .336 .296

C. 2007 1 2 3 4 5

Average density .007 (.004)

.002 (.005)

.019** (.004)


.061** (.016)

.085** (.013)

Unemployment rate

-.010** (.002)

-.006* (.003)

-.007** (.002)

-.010** (.002)

Income .055* (.027)

.128** (.022)

Adjusted R2 .603 .624 .509 . 585 .427

Notes: The table summarizes the estimation results of (8) in the main text by year. Heteroskedastic-consistent standard errors are provided in parentheses underneath the corresponding estimates. ** and * indicate statistical significance at the 1 and 5 % levels, respectively.

Table A-4-1. Density effects on relative sectoral prices – the 1997 sample

1 2 3 4 5 A. Industrial products

Average density

.046** (.009)

.022† (.013)

.081** (.010)


.205** (.039)

.356** (.047)

Unemployment rate -.015† (.008)

-.003 (.010)

-.006 (.010)

-.015 (.010)

Income .200* (.078)

.442** (.086)

Adjusted R2

.684 .706 .595 .599 .385

B. Goods

Average density

.048** (.009)

.025† (.013)

.081** (.011)


.198** (.038)

.353** (.049)

Unemployment rate -.017† (.008)

-.004 (.011)

-.007 (.011)

-.017 (.011)

Income .194* (.075)

.444** (.084)

Adjusted R2

.700 .721 .614 .604 .407

Notes: The table summarizes the estimation results of (9) in the main text and its variant specifications for the 1997 data. Heteroskedastic-consistent standard errors are provided in parentheses underneath the corresponding estimates. ** , * and † indicate statistical significance at the 1 , 5 and 10 % levels, respectively.

Table A-4-2. Density effects on relative sectoral prices – the 2002 sample 1 2 3 4 5 A. Industrial products

Average density

.025** (.008)

.012 (.010)

.054** (.008)


.165** (.026)

.246** (.028)

Unemployment rate -.026** (.004)

-.017* (.007)

-.017** (.006)

-.027** (.005)

Income .120* (.048)

.309** (.058)

Adjusted R2

.743 .764 .593 .681 .428

B. Goods

Average density

.026** (.008)

.014 (.011)

.055** (.009)


.167** (.025)

.251** (.030)


-.019* (.007)

-.019** (.006)

-.029** (.005)

Income .116* (.049)

.317** (.061)

Adjusted R2

.765 .784 .612 .699 .445

Notes: The table summarizes the estimation results of (9) in the main text and its variant specifications for the 2002 data. Heteroskedastic-consistent standard errors are provided in parentheses underneath the corresponding estimates. ** and * indicate statistical significance at the 1 and 5 % levels, respectively.

Table A-4-3. Density effects on relative sectoral prices – the 2007 sample 1 2 3 4 5 A. Industrial products

Average density

.020* (.009)

.008 (.010)

.038** (.007)

DID density .088* (.037)

.089** (.032)

.155** (.028)


-.004 (.004)

-.009** (.002)

-.014** (.003)

Income .129* (.048)

.228** (.049)

Adjusted R2

.582 .624 .524 .532 .419

B. Goods

Average density

.021* (.008)

.009 (.010)

.041** (.007)


.100** (.031)

.169** (.029)


-.009* (.004)

-.013** (.002)

-.019** (.003)

Income .125* (.046)

.260** (.048)

Adjusted R2

.661 .696 .592 .610 .489

Notes: The table summarizes the estimation results of (9) in the main text and its variant specifications for the 2007 data. Heteroskedastic-consistent standard errors are provided in parentheses underneath the corresponding estimates. ** and * indicate statistical significance at the 1 and 5 % levels, respectively.

Appendix A3. Penn Effects Among Rich and Poor Prefectures

Some studies (Kravis and Lipsey, 1987; Cheung, Chinn, and Fujii, 2007; Fujii, 2013b)

suggest that the income effect on prices could vary between economies at different stages of

development and is stronger for developed economies than for developing ones. Since the

stage of development is closely related to the level of income, the Penn effect tends to be

more substantial among higher income countries than lower income countries. Does a similar

regularity hold in the intra-Japan context?

To investigate the issue, we rank the forty-seven Japanese prefectures according to

their per capita income and construct two subsamples – the upper and lower income groups

whose income levels are, respectively, above and below the median level. The Penn effect

regression (5) was re-run on each of these two subsamples. The year-by-year ̂ s graphed in

Figure A-1 clearly show that the Penn effect is mainly an upper income group phenomenon.

The ̂ s from the upper income group are larger than those from the lower income group and

those in Figure 1. More importantly, the upper income group estimates are statistically

significant while the lower income group ones are not.

Figures A-2 and A-3 plot the year-by-year estimates, ̂ s, from the upper and lower

income group samples constructed in a similar fashion using the PWT (version 7.0) and the

WDI, respectively. While the numerical values are different from those in Figure A-1, the

qualitative results pertaining to the upper and lower income groups are the same. The upper

income group yields significant and large income effect estimates while the lower income

group obtains insignificant and small estimates.

In sum, comparable to the results from the cross-country data, the significant income

effect on prices displayed by the Japanese data appears to be driven mainly by rich, rather

than poor, prefectures.

Figure A-1. Penn effects for the high and low income Japanese Prefectures

Notes: The figure plots the Penn effect coefficient estimates and their p-values of the high and low income Japanese prefectures sub-samples obtained from (5) in the main text.

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

Upper income (coefficient) Lower income (coefficient) Upper income (p-value) Lower income (p-value)

Figure A-2. Penn effects for high and low income sub-samples of the PWT 7.0 data

Notes: The figure plots the Penn effect coefficient estimates and their p-values of the high and low income sub-samples of the PWT 7.0 data obtained from (5) in the main text.

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

Upper (coefficient) Upper (p-value) Lower (coefficient) Lower (p-value)

Figure A-3. Penn effects for high and low income sub-samples of the WDI data

Notes: The figure plots the Penn effect coefficient estimates and their p-values of the high and low income sub-samples of the WDI 7.0 data obtained from (5) in the main text.

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

Upper (coefficient) Upper (p-value) Lower (coefficient) Lower (p-value)

Reference

Asea, P. and Mendoza, E. (1994) The Balassa-Samuelson model: A general equilibrium

appraisal, Review of International Economics, 2, 244-267.

Balassa, B. (1964) The purchasing power doctrine: A reappraisal, Journal of Political

Economy, 72, 584-96.

Bergin, P.R. (2009) Balassa-Samuelson effect, in K.A. Reinert and R.S. Rajan (eds.) The

Princeton Encyclopedia of the World Economy Volume 1, Princeton University Press,

Princeton, NJ.

Bergstrand, J.H. (1991) Structural determinants of real exchange rates and national price

levels: Some empirical evidence, American Economic Review, 81, 325-34.

Bhagwati, J.N. (1984) Why are services cheaper in the poor countries? Economic Journal 94,

279-86.

Canzoneri, M.B., Cumby, R.E., and Diba, B. (1999) Relative labor productivity and the real

exchange rate in the long run: Evidence for a panel of OECD countries, Journal of

International Economics, 47, 245-66.

Carlino, G.A. and Voith, R. (1992) Accounting for differences in aggregate state productivity,

Regional Science and Urban Economics, 22, 597-617.

Chen, S. and Ravallion, M. (2010a) China is poorer than we thought, but no less successful in

the fight against poverty, in S. Anand, P. Segal, and J.E. Stiglitz (eds.) Debates on the

Measurement of Global Poverty, Oxford University Press, New York, NY.

Chen, S. and Ravallion, M. (2010b) The developing world is poorer than we thought, but no

less successful in the fight against poverty, Quarterly Journal of Economics, 125,

1577-625.

Cheung, Y.-W., Chinn, M., and Fujii, E. (2007) The overvaluation of renminbi undervaluation,

Journal of International Money and Finance, 26, 762-785.

Cheung, Y.-W., Chinn, M., and Fujii, E. (2009) Pitfalls in measuring exchange rate

misalignment: The yuan and other currencies, Open Economies Review, 20, 183-206.

Chinn, M. D. (2000) The usual suspects? Productivity and demand shocks and Asia-Pacific

real exchange rates, Review of International Economics, 8, 20-43.

Ciccone, A. and Hall, R.E. (1996) Productivity and the density of economic activity,

American Economic Review, 86, 54-70.

Ciccone, A. and Jarocinksi, M. (2010) Determinants of economic growth: Will data tell?

American Economic Journal: Macroeconomics, 2, 223–47.

Deaton, A. and Heston, A. (2010) Understanding PPPs and PPP-based national accounts,

American Economic Journal: Macroeconomics, 2, 1–35.

De Gregorio, J., Giovannini, A., and Wolf, H.C. (1994) International evidence on tradable

and nontradable inflation, European Economic Review, 38, 1225-44.

Engel, C. and Rogers, J.H. (1996) How wide is the border? American Economic Review, 86,

1112-25.

Froot, K.A. and Rogoff, K. (1991) The EMS, the EMU, and the transition to a common

currency, in S. Fisher and O. Blanchard (eds.) National Bureau of Economic Research

Macroeconomics Annual, MIT Press, Cambridge, MA.

Fujii, E. (2013a) The Penn effect: Decoupling and re-coupling in the price-income

relationship? in Y.-W. Cheung and F. Westermann (eds.) Global Interdependence,

Decoupling, and Re-coupling, MIT Press, Cambridge, MA.

Fujii, E. (2013b) Reconsidering the Price-Income Relationship across Countries. CESifo

Working Paper 4129.

Glaeser, E.L. (2008) Cities, Agglomeration and Spatial Equilibrium, Oxford University Press,

New York, NY.

Glaeser, E.L. and Maré, D.C. (2001) Cities and skills, Journal of Labor Economics, 19,

316-42.

Henderson, J.V. (1974) The sizes and types of cities, American Economic Review, 64, 640-56.

Hsieh, D.A. (1982) The determination of the real exchange rate: The productivity approach,

Journal of International Economics, 12, 335-62.

Imbs, J.M., Mumatz, H., Ravn, M.O., and Rey, H. (2010) One TV, one price? Scandinavian

Journal of Economics, 112, 753-81.

Johnson, S., Larson, W., Papageorgiou, C. and Subramanian, A. (2009) Is newer better? The

Penn World Table revisions and the cross-country growth literature, NBER Working

Paper 15455.

Kakkar, V. (2003) The relative price of nontraded goods and sectoral total factor productivity:

An empirical investigation, Review of Economics and Statistics, 85, 444-52.

Kravis, I.B., Heston, A., and Summers, R. (1978) International Comparisons of Real Product

and Purchasing Power, Johns Hopkins University Press, Baltimore, MD.

Kravis, I.B. and Lipsey, R.E. (1983) Toward an explanation of national price levels,

Princeton Studies in International Finance No. 52, Department of Economics,

Princeton University, Princeton.

Kravis, I.B. and Lipsey, R.E. (1987) The assessment of national price levels, in S.W. Arndt

and J. D. Richardson (eds.) Real Financial Linkages among Open Economies, MIT

Press, Cambridge, MA.

Krugman, P.R. (1991) Increasing returns and economic geography, Journal of Political

Economy, 99, 483-99.

Marston, R.C. (1987) Real exchange rates and productivity growth in the United States and

Japan, in S.W. Arndt and J. D. Richardson (eds.) Real Financial Linkages among

Open Economies, MIT Press, Cambridge, MA.

Milanovic, B. (2009) Global inequality recalculated: The effect of new 2005 PPP estimates

on global inequality, World Bank Policy Research Working Paper 5061.

Officer, L.H. (1976) The productivity bias in purchasing power parity: An econometric

investigation, International Monetary Fund Staff Papers, 23, 545-79.

Ponomareva, N. and Katayama, H. (2010) Does the version of the Penn World Tables matter?

An analysis of the relationship between growth and volatility, Canadian Journal of

Economics, 43, 152-79.

Samuelson, P.A. (1964) Theoretical notes on trade problems, Review of Economics and

Statistics, 46, 145-54.

Samuelson, P.A. (1994) Facets of Balassa-Samuelson thirty years later, Review of

International Economics, 2, 201-26.

Summers, R. and Heston, A. (1991) The Penn World Table (Mark V): An extended set of

international comparisons, 1950-1988, Quarterly Journal of Economics, 104, 51-66.

Table 1. Penn effect by product category

1997 2002 2007 Pooled CPI 0.138**

(0.029) 0.100** (0.029)

0.128** (0.022)

0.122** (0.015)

CPI excluding fresh foods 0.094** (0.020)

0.073** (0.022)

0.105** (0.016)

0.092** (0.010)

Goods 0.025

(0.023) 0.047

(0.030) 0.040

(0.029) 0.038* (0.015)

Agricultural & aquatic products 0.154** (0.040)

0.174** (0.039)

0.066 (0.049)

0.127** (0.025)

Fresh agricultural & aquatic products

0.170** (0.048)

0.195** (0.044)

0.069 (0.054)

0.140** (0.028)

Industrial products 0.028 (0.026)

0.055 (0.033)

0.072* (0.033)

0.053** (0.017)

Publications 0.039 (0.029)

0.054 (0.033)

0.001 (0.025)

0.029 (0.016)

Electricity, gas & water charges -0.234** (0.058)

-0.204** (0.061)

-0.197** (0.058)

-0.211** (0.032)

Services 0.352**

(0.069) 0.229** (0.062)

0.185** (0.049)

0.252** (0.034)

Public services 0.032 (0.036)

-0.009 (0.019)

0.015 (0.014)

0.013 (0.013)

General services 0.470** (0.094)

0.365** (0.086)

0.300** (0.074)

0.374** (0.046)

Private house rent 1.341** (0.256)

0.957** (0.270)

0.864** (0.244)

1.044** (0.140)

Eating out 0.221** (0.059)

0.159** (0.035)

0.085** (0.024)

0.151** (0.024)

Notes: The table summarizes the estimation results of (7) in the main text. Heteroskedastic-consistent standard errors are provided in parentheses underneath the corresponding estimates. ** and * indicate statistical significance at the 1 and 5 % levels, respectively. Year-specific constants are allowed for the pooled data estimation.

Table 2. Density effects on the price level – the pooled estimates 1 2 3 4 5

Average density

0.009** (0.002)

0.004 (0.003)

0.020** (0.002)

DID density 0.061** (0.009)

0.061** (0.008)

0.092** (0.008)

Unemployment rate -0.009** (0.001)

-0.005** (0.002)

-0.006** (0.001)

-0.009** (0.001)

Income 0.051** (0.018)

0.122** (0.015)

Adjusted R2

0.600 0.623 0.492 0.555 0.386

Notes: The table summarizes the estimation results of (8) in the main text and its variant specifications using sample of the 1997, 2002, and 2007 data. Year-specific constants are included in all specifications. The number of observations is 141. Heteroskedastic-consistent standard errors are provided in parentheses underneath the corresponding estimates. ** and * indicate statistical significance at the 1 and 5 % levels, respectively.

Table 3. Correlations between explanatory variables

Income Average density DID density

Average density 0.58

DID density 0.33 0.77

Unemployment rate -0.36 0.28 0.42

Notes: The table gives correlation coefficient estimates between explanatory variables considered in the main text. These estimates are computed using data pooled from the 1997, 2002, and 2007 samples.

Table 4. Density effects on relative sectoral prices – the pooled estimates 1 2 3 4 5 A. Industrial products

Average density

0.031** (0.006)

0.011 (0.007)

0.056** (0.005)

DID density 0.140** (0.023)

0.140** (0.022)

0.240** (0.022)


-0.002 (0.004)

-0.008** (0.003)

-0.016** (0.003)

Income 0.186** (0.036)

0.321** (0.039)

Adjusted R2

0.598 0.642 0.520 0.528 0.372

B. Goods

Average density

0.033** (0.006)

0.013 (0.006)

0.058** (0.005)

DID density 0.143** (0.021)

0.143** (0.020)

0.248** (0.022)


-0.006 (0.004)

-0.011** (0.003)

-0.019** (0.003)

Income 0.179** (0.034)

0.336** (0.037)

Adjusted R2

0.633 0.674 0.552 0.557 0.409

Notes: The table summarizes the estimation results of (9) in the main text and its variant specifications using sample of the 1997, 2002, and 2007 data. Year-specific constants are included in all specifications. The number of observations is 141. Heteroskedastic-consistent standard errors are provided in parentheses underneath the corresponding estimates. ** and * indicate statistical significance at the 1 and 5 % levels, respectively.

Figure 1. Time profile of the Penn effect within Japan

Notes: The figure plots the Penn effect coefficient estimates and their p-values of the Japanese data obtained from (5) in the main text.

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 20080

0.0005

0.001

0.0015

0.002

0.0025

Penn coefficient (left scale) P-value (right scale)

Figure 2. The international and intra-Japan Penn effects

Notes: The figure plots the Penn effect coefficient estimates and their p-values of a) the international PWT7.0 and WDI (January 2012) data and b) the Japanese data obtained from (5) in the main text.

0

0.05

0.1

0.15

0.2

0.25

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

Intra-Japan PWT7.0 WDI

Figure 3. Income effects on the Japanese disaggregated price indexes

Notes: The figure plots the estimates of income effect on price indexes of disaggregated product categories obtained from (7) in the main text.

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6CP

I

Goo

ds

Agr

icul

tura

lpr

oduc

ts

Fres

hag

ricul

tura

lpr

oduc

ts

Indu

stria

lpr

oduc

ts

Publ

icat

ions

Elec

trici

ty,

gas &

wat

er

Serv

ices

Publ

icse

rvic

es

Gen

eral

serv

ices

Hou

se re

nt

Eatin

g ou

t

1997 2002 2007

Documents

Research Centre for International Economics Working Paper: … working paper... · 2014. 6. 9. · Research Centre for International Economics . Working Paper: 2014008 . Title of